Do people like you more than you think? Usually, yes

The most direct measurement of this comes from a set of studies published in 2018. Researchers paired strangers for a conversation and then asked each person two things: how much they liked their partner, and how much they thought their partner liked them. People consistently guessed too low. Their partners liked them more than they believed.

The same team then followed students who shared university housing across an academic year, and ran the comparison with adults in workshops. The gap showed up again, and among the students it lasted for months. That is the part worth sitting with. A misjudgement at a first meeting would be unremarkable. One that survives living alongside someone is a stable distortion in how people read the way they land on others.

The limits are real. The samples were mostly students and workshop groups, from the cultures where this has been studied most; liking was self-reported on both sides; and the size of the gap varied considerably between studies. What did not vary was the direction. Nobody found the opposite pattern, where people reliably think they were liked more than they were.

One likely reason for the gap is asymmetric access. During a conversation you hear your own commentary in full: that joke did not land, you talked too long about work, you lost the thread of their story. You hear none of theirs. They were having their own version of that commentary, about themselves, and none of it was about you.

So when you replay the conversation afterwards, you are reviewing a recording with only one microphone. Every awkward pause is audible and every doubt is attributed to the other person, because there is nowhere else for it to go. The other person, meanwhile, mostly remembers whether the conversation felt warm and whether you seemed interested in them.

A related finding points the same way. In experiments that assigned people to deep or shallow conversations, people predicted the deep ones would be more awkward and less connecting than they turned out to be, and they were most wrong about the deepest topics. One proposed mechanism is that people underestimate how interested others are in them. Those conversations were assigned, often between strangers who knew they were in a study, so the stakes were lower than in life. The forecast still leaned the same pessimistic way.

The relationship literature gives a working answer to what people notice. Intimacy, operationally, is feeling understood, validated and cared for, and what predicts the behaviours that keep a relationship going is perceived responsiveness: the other person’s sense that you were paying attention to them. It is a framework resting on mostly correlational work, so it describes what goes together rather than what causes what. But it is a useful correction to the replay, which grades you on fluency and wit, while the person across from you was registering whether you listened.

Research on conversation reaches a similar place. In two laboratory experiments and a speed-dating study, people who asked more questions, follow-up questions in particular, were liked more, and in speed dating they were more often offered a second date. That study carries two caveats: one co-author has faced well-publicised data-integrity allegations, and this specific effect has been contested and defended in published comment and reply. It is corroboration, not the argument. The argument is the responsiveness work, which points the same way.

And if what you are replaying is that you only made small talk: a pooled study of 486 people, using recorders that sampled real conversations from ambient audio, found that small talk was unrelated to wellbeing in either direction, while more substantive conversation went with higher life satisfaction. It is correlational and cannot say which came first. It does say that an easy, light conversation is not a failure.

If one conversation feels like a verdict, it helps to know how little a single conversation decides. In a classroom study from 1992, four women of similar appearance attended a large lecture course 0, 5, 10 or 15 times and never spoke to anyone. At the end of term, 130 students rated them from photographs, and they liked the woman who had been there most substantially more, despite never having talked to her.

That is one study, with one course, and students rating slides rather than people they knew, so it cannot carry much on its own. The general effect it demonstrates, that familiarity raises liking, is well replicated. The practical point is that liking is not settled at the first meeting. It accumulates with being around.

The same pessimism shows up at the next step, when you think about getting in touch. In a series of experiments, people who sent an unprompted message predicted how the recipient would feel, and recipients appreciated it more than the senders expected. These were single moments, not friendships followed over time, and nobody has shown that acting on the finding makes a friendship last. But it means the hesitation before the follow-up message is the same forecast as the replay, with the same lean.

When the replay starts, name it as what it is: a review of a conversation with half the evidence missing. You know your own doubts in detail and nothing of theirs, and the research says that leaves people guessing too low about how they came across. That does not mean every conversation went well. It means the replay is a poor instrument for finding out.

Better evidence comes from the next contact, not from rewinding the last one. Send the follow-up message, and put one specific thing in it from what they told you: the trip they were planning, the book they mentioned. A remembered detail is one of the clearest signals that you were listening, which is the thing they were noticing in the first place, and it gives them something easy to reply to.

None of this needs an app. Jot down one thing the person said before the conversation fades, and use it the next time you write or see them. What does not help is going back over the conversation looking for the moment it went wrong, because that search runs on the one-sided evidence the liking gap was measured against.